OpenAI | Evaluating fairness in ChatGPT

This study analyzed how ChatGPT responds based on users’ names, investigating potential biases related to cultural, gender, and racial associations. The research found that name-based biases affected responses in less than 0.1% of cases. Older models showed higher rates of gender-related biases, but newer models have significantly reduced these rates.

Source: Here


Comments

Leave a Reply

Your email address will not be published. Required fields are marked *